What if it all goes right?
We bet our company on AI. Here’s the case for optimism.
In 2022, Intercom (now Fin) faced a reckoning. Five consecutive quarters of declining revenue growth required serious action. Our founding CEO Eoghan returned with a five-point plan, we refocused our efforts around a single product category, and underwent a reduction in force. And then, ChatGPT arrived. Our co-founder Des Traynor wondered if the emergence of AI was “adding insult to injury, or the single scenario that we can’t survive?”
All around us, SaaS companies were getting “sold off for parts to private equity.” The end seemed near.
But after a weekend of experimenting with the newly released LLM, the future appeared less bleak than initially forecast. As Des put it: “AI might actually be both the death of our company and also the best chance we have of survival.”
By Monday, we decided to go all in, and over the next three years we rebuilt almost everything we’d spent the previous eleven years building, including the name.
Des recently talked through how that happened – on The Further, Faster Podcast and the AI Revolution Show. What follows pulls the two conversations together: the entrepreneurial lessons learned along the way, what to consider when undergoing your own AI pivot, and why Des thinks we’ve grown too pessimistic about the AI-powered future.
Focus on the product in front of you
Before the AI story, there’s the unglamorous part that made it possible. On The Further, Faster Podcast, Des walks us through it and what he thinks today’s entrepreneurs should focus on.
Firstly, the job of a founder is “to do something that doesn’t make sense up until the point that it does make sense” – and the moment it makes sense, the founder phase is over and you’re running a business. Des feels it’s not good to idolize or obsess over the label of founder because it should be a temporary role; don’t cling to it rather than admitting your company has stopped working. The entrepreneurs he’d back again are the ones who “closed their business down, gave the money back, took a couple of years working in a good company, got some new ideas, and went again.”
That’s also why he’s wary of early-stage companies chasing investment capital. “A really long runway is not what you should be optimizing for. You should be optimizing for success.” Chase the valuation and you end up watching your own stock price instead of asking the only questions that matter: “are we building good software? Do people like it? Are they paying a good amount of money for it?”
You can feel the difference in the products themselves. Des refers to the experience of building Exceptional – an error-tracking tool the team developed before Intercom – as “doing your best to try and push it up a hill”. Intercom was the opposite: it “was a very easy product to bring out,” viral in the way B2B occasionally is. Early users routinely praised it, even just a few months in.
This is why it’s important to ship. Founders often delay launching because they’re afraid to find out their product doesn’t work, and a reputation as a “considerate, tasteful founder” is a comfortable disguise for that fear. After you ship, take user feedback seriously but not blindly. When they ask for 11 more features, don’t assume they’re right – “that could be their way of saying this is a bag of shit.”
Becoming an AI company takes sacrifice
The decision to go all in on AI was, oddly, the easy part. It was clear, customer service was about to be changed completely, and it was a race against the clock to be the first to market.
The hard part was everything after the decision. A transformation like this touches the whole company – the branding, the pricing, what you sell, how you build, and who you compete with. On the AI Revolution Show, Des compares it to the ship of Theseus: change enough of the parts and it’s hard to say what’s still the original boat. “Maybe the canteen is the same,” he joked.
What made that survivable was culture set deliberately, in advance. When Eoghan returned and rewrote the company values, three of them – resilience, open-mindedness, and impatience – read, in hindsight, like preparation for exactly this. So when we made radical decisions like breaking up teams or abandoning roadmaps, the ground had already been laid.
For companies going through change, Des also warns not to “be a prisoner of nostalgia.” Intercom was a longstanding, beloved brand. On the surface, sacrificing it looks like a significant risk. But he argues the opposite. “Being known isn’t always useful – it can almost be a counterweight.”
He tested this theory with a thought experiment: “If I said the coolest new payroll app is built by Workday, you’d be like, really?” The name you’ve spent a decade building can be the very thing that stops people believing you’re building something new, so you have to be willing to “shred and shed that brand legacy.”
Building and selling the AI-native way
Plenty of products call themselves AI-native. Most fail the first test: does the product work? Customer support is load-bearing – if it breaks, the business breaks – so “it works most of the time” isn’t good enough.
Clearing that bar takes more than a wrapper. We’ve gone down to the model layer with Apex, and what looks from the outside like a single product is, in Des’s words, “more like 27 systems connected together,” several revisions deep.
Looks can be deceiving though. It’s common for users to think they can recreate Fin with a frontier model, and maybe they can reach a 30 or 40% resolution rate in two minutes. “The other forty percent is going to take you three years and 60 people.” We watch prospects try to build their own, get stuck, and come back – and sometimes the only useful thing is to let them “bang their head off a brick wall” and leave the door open.
Building for AI is different from building software. We used to start with customer research, but because customers aren’t experts in what AI now makes possible, you now have to start with what’s technically reliable and take that to the market. It’s a reversal Des expects to be temporary, until the technology settles.
You also have to rethink design. The AI-native instinct isn’t to begin with “boxes and arrows and buttons.” It’s to take the least amount from the user and give back the most. The example Des provides is bug reporting. Rather than relying on users to report issues, a bug should be automatically detected so the experience goes “from seven screens to no screens.”
And the impact of all of this must be easily demonstrable. According to Des: “brand will get you the demand, but ultimately performance is what gets you the contract.” Serious buyers run bake-offs – a thousand of their own real questions across three vendors – and we lean into it. Rigour favours whoever is actually best; deliver results and the sales will follow.
What if it all goes right?
AI is going to keep moving. It’s a realization Des has had to come to terms with: “I’ve probably been more wrong than right whenever I’ve said … I think the AI is settled here.”
What does this pace of advancement mean for the wider market? The walls between products will come down – point solutions will be absorbed by whoever owns the rest of the workflow. Des thinks the systems of record will endure, and a smaller number of larger businesses will dominate.
It’s reasonable to feel apprehension towards such a mammoth change, it’s a narrative that dominates the media. And overall, the conversation about AI has become fluent in every way it might go wrong. But we’re strangely quiet about the other possibility. “No one’s asking what if it all goes right,” Des says – what it might mean for things like education or healthcare access.
The near term will hold real dislocation, and we’re not pretending otherwise. But the upside, he argues, “is potentially the biggest thing we’re ever going to see for humanity.”
We’ve already run the small version of that experiment on ourselves. We bet the company, disrupted the thing we’d built, and came out the other side stronger for it. That isn’t proof the larger bet pays off. But it’s reason enough to take the optimistic question seriously – and to keep building as though the answer is yes.
Watch or listen to the full conversations: Des Traynor on The Further, Faster Podcast and the AI Revolution Show.